Best AI Twitter (X) Accounts to Follow in 2026
ZarifAI Twitter—now AI X—can alert you to a release or paper quickly. It can also turn an unverified screenshot into a consensus before anyone opens the underlying artifact.
Updated September 16, 2026 — handles and reasons verified.
Last verified September 16, 2026. Recheck target: monthly. Download CSV (30 entries).
AI Twitter or AI X is the network of researchers, engineers, founders, educators, policymakers, and official organizations discussing artificial intelligence on X. A useful account links to papers, code, documentation, evaluations, or firsthand experiments instead of merely repeating news.
The goal is not to follow the most accounts. Build a balanced feed from people and organizations that link primary artifacts, document experiments, show expertise, correct mistakes, and cover different parts of the AI ecosystem. Linking primary artifacts matters most; the rest just help you judge what they link.
Start with five
These five earn a follow before you add a lab account or a commentator.
- Andrej Karpathy (@karpathy) · A few times a week, in bursts
Longer explanations, talks, and experiments that build LLM intuition instead of repeating launch copy.
- Simon Willison (@simonw) · Daily when something new ships
Documented prompts, outputs, and linked notes when a new model or API ships.
- Ethan Mollick (@emollick) · Several times a week
Research-informed guidance on AI at work and in education, with links to the longer writing.
- Sebastian Raschka (@rasbt) · Several times a week
Diagrams, code, and paper discussion that show what actually changed in a model.
- Chip Huyen (@chipro) · A few times a week
Evaluation, data, latency, cost, and the gap between a demo and a system you can run.
The directory
The matrix is the complete shortlist. Use the roles deliberately: research for papers, engineering for code and operating tradeoffs, primary sources for launch links, policy and public interest for the launch cycle's blind spots.
Scroll horizontally to compare the columns.
| Name | Role | Why it is here | Cadence |
|---|---|---|---|
| Andrej Karpathy @karpathy | Research explainer | Longer explanations, talks, and experiments that build LLM intuition instead of repeating launch copy. | A few times a week, in bursts |
| Simon Willison @simonw | Hands-on testing | Documented prompts, outputs, and linked notes when a new model or API ships. | Daily when something new ships |
| Ethan Mollick @emollick | Applied use | Research-informed guidance on AI at work and in education, with links to the longer writing. | Several times a week |
| Sebastian Raschka @rasbt | Technical explanation | Diagrams, code, and paper discussion that show what actually changed in a model. | Several times a week |
| Chip Huyen @chipro | Engineering systems | Evaluation, data, latency, cost, and the gap between a demo and a system you can run. | A few times a week |
| Demis Hassabis @demishassabis | Research | Primary research announcements from Google DeepMind, especially AI-for-science. | When DeepMind ships |
| Fei-Fei Li @drfeifei | Research | Vision, spatial intelligence, and human-centered AI from someone who still publishes. | When there is research or institutional news |
| François Chollet @fchollet | Research | Reasoning, ARC, and evaluation arguments that survive a launch cycle. | A few times a week |
| David Ha @hardmaru | Research | Generative and world-model papers with visual experiments attached. | Several times a week |
| Jim Fan @DrJimFan | Research | Robotics and embodied-AI research explainers that link the paper or demo. | Several times a week |
| Nathan Lambert @natolambert | Research | Open models and post-training commentary from someone who trains the models. | Daily-ish |
| Andrew Ng @AndrewYNg | Applied AI | Accessible industry and education perspective without pretending every launch is a phase change. | A few times a week |
| swyx @swyx | Engineering | AI-engineering tools, events, and implementation patterns from the Latent Space orbit. | Daily |
| Hamel Husain @HamelHusain | Engineering | Evaluation and LLM-engineering methods with the failure modes included. | Several times a week |
| Shreya Shankar @sh_reya | Engineering | Data systems and evaluation thinking for applications that have to stay correct. | A few times a week |
| Harrison Chase @hwchase17 | Engineering | Primary updates from the agent-framework ecosystem, to be checked against changelogs. | Several times a week |
| OpenAI @OpenAI | Primary source | Official announcement links for OpenAI releases; still an organization's perspective. | When they ship |
| Anthropic @AnthropicAI | Primary source | Official announcement links for Claude, research, and policy posts. | When they ship |
| Google DeepMind @GoogleDeepMind | Primary source | Research and product links from Google DeepMind's own account. | When they ship |
| AI at Meta @AIatMeta | Primary source | Official Meta AI research and open-model links. Not @MetaAI. | When they ship |
| Mistral AI @MistralAI | Primary source | Official model and product links from Mistral. | When they ship |
| Hugging Face @huggingface | Primary source | Repository, model, and community links for the open-model ecosystem. | Daily |
| NVIDIA AI @NVIDIAAI | Primary source | Developer and research links for inference, systems, and platforms. | Several times a week |
| Cohere @Cohere | Primary source | Official product and research links for enterprise LLM work. | When they ship |
| Perplexity @perplexity_ai | Primary source | Official product links; treat as marketing until you open the underlying change. | When they ship |
| Stanford HAI @StanfordHAI | Policy | Institutional research and policy links that sit outside the launch cycle. | A few times a week |
| NIST @NIST | Policy | Official technical guidance on measurement, standards, and AI risk. | When NIST publishes |
| OECD Innovation @OECDinnovation | Policy | Cross-country policy and data rather than lab marketing. | When OECD publishes |
| Ada Lovelace Institute @AdaLovelaceInst | Public interest | Governance and social-impact research that names who is affected. | When they publish |
| AI Now Institute @AINowInstitute | Public interest | Accountability and labor analysis that counterbalances launch-week consensus. | When they publish |
How this list is maintained
Accounts stay when they have a clear job in a reader's feed. Follower counts are ignored. Recheck activity, handle changes, and original-source links each month.
A practical starting setup is four private X Lists, not one algorithmic home feed: primary sources, research, engineering, and work or policy. Start small—about five to eight accounts per list. X is a discovery layer: open the original paper, code, documentation, evaluation, or long-form post before acting on a claim.
Keep an account when at least two of these are true: it regularly links primary sources; it adds expertise you cannot get from a lab announcement; it shows methods, prompts, code, or limitations; it separates fact, interpretation, and prediction; it corrects earlier claims. Mute accounts whose feed is mostly outrage, affiliate promotion, or screenshots without context.
For a source layer beyond social posts, use AI blogs and news sites alongside this feed. Discussion that needs to stay findable belongs in AI communities, not in a 24-hour reply thread.
Want a weekly source-controlled reading list in addition to your X feed? Build a weekly AI article recommendation workflow that collects, deduplicates, and ranks the sources you choose.
See the change log below for updates to this directory.
Change log
Moved onto the directory template: last-verified date, start-with-five, per-entry reason and cadence, and a CSV download. Handles rechecked.
Links and affiliations verified; Meta AI handle updated to @AIatMeta.